Microsoft launches Microsoft-Decision-1, a fast decision-scoring model, on Foundry and OpenRouter
Microsoft released Microsoft-Decision-1, a model that returns calibrated probabilities over fixed options instead of generating text. Microsoft says it topped a 36-benchmark comparison and ran 35 times faster than GPT-6 Sol.
Microsoft on October 9 introduced Microsoft-Decision-1, a new model for fast decision-scoring, available in Microsoft Foundry and through OpenRouter, the company announced on its Command Line blog.
A model that decides rather than writes
Unlike large language models designed to generate text or reason through complex problems, decision models are built to deliver structured outputs that software can act on immediately. Given a fixed set of answer options, Decision-1 returns a calibrated probability score for each one. It supports yes/no, multiple-choice and rating questions, as well as rubric-based grading of AI responses and agent actions, and is aimed at routing, classification, prioritization, verification and workflow control.
Microsoft built the model by post-training Alibaba's open-weight Qwen3.5-9B and said it will soon rebase it on other models, including Microsoft AI (MAI) and OpenAI models.
Microsoft's reported results
- Highest accuracy in a 36-benchmark comparison spanning nearly 150,000 questions kept blind from training (83.5%, according to The Register).
- Median (P50) latency 2.5 times faster than runner-up H2O-Lightning-4B v1.1 and 35 times faster than OpenAI's GPT-6 Sol.
- Decisions flipped on just 1.3% of perturbed requests on average across eight perturbation types.
- Safety testing on 5,250 requests across 11 benchmarks covering harmful content, jailbreaks and prompt injection.
Input tokens cost $0.042 per million, with output free, The Register reported.
A crowded new category
The Register noted that the category has grown rapidly since TypeSafe AI unveiled its Jev decision model, with OpenAI, Cloudflare, Perplexity, Snowflake, H2O.ai and others announcing rival offerings and more than 100 such models now competing. "Now that agentic AI is a reality, we've seen that cost plays a major role in how people decide to use AI," said Achint Srivastava, VP of software engineering in Microsoft's Office of the CTO.
Sources: Microsoft Command Line blog; The Register.
Related articles
Anthropic adds live dashboards and animated explainers to Claude with Dashboards and Motion
Anthropic launched Claude Dashboards, which builds auto-updating dashboards from data platforms and CRMs, and Claude Motion, which turns reports and charts into editable animations, in beta, while making Docs, Slides and Design available on every plan.
Microsoft’s Satya Nadella says advanced AI needs a human-controlled ‘emergency brake’
Microsoft CEO Satya Nadella urged the industry to treat frontier AI models like insider risks, wrapping them in containment, independent controls and an “emergency brake” that lets an authorized person pause or shut down a model mid-task.
Amazon in talks to buy AI startup Decart for around $7 billion, WSJ reports
Amazon is negotiating to acquire Decart, maker of real-time video-editing and robotics-simulation AI models, in a deal that could be valued at about $7 billion, The Wall Street Journal reported. Decart was valued at nearly $4 billion in May.
A college in Illinois tried to cancel its Flock contact and it refused
In April of 2025, Elgin Community College had six Flock safety cameras installed to "enhance the safety and security of students.
Silicon Valley is falling in line with Trump’s attempt to rename AI
When President Donald Trump announced his plans to rebrand artificial intelligence as "super intelligence," you'd be forgiven for being skeptical that it would catch on.
The new Apple TV could have a built-in mic and a remote with a speaker
The latest rumors are that the new Apple TV will have a built-in microphone, while the upgraded remote will get a speaker.